4.3 Estimating mortality and number of deaths due to AMR
Table 1 contains information on the data sources available for estimating the mortality and number of deaths due to AMR as well as their strengths and limitations.
Note on choosing comparators for community-origin AMR infections
An important (and often neglected) consideration when estimating deaths due to community-origin AMR infections is the choice of data source from which the comparator cohort comes and represents.
For example, if the aim is to calculate excess deaths due to community-origin AMR compared with the AMR infection-free counterfactual (see definition in the previous section), it is important to note that patients admitted to hospital without community-origin infections will not be an appropriate comparator. This is because patients admitted to hospital for other conditions (such as stroke, surgery, accidents, etc.) may be expected to have a higher mortality risk than the general population in this community. Instead, you would need a cohort of individuals from the same community as those with community-origin AMR infections but without bacterial infections.
| Measurements | Data sources | Gaps and limitations |
|---|---|---|
| Crude mortality | Death registry data Hospital clinical data International Statistical Classification of Disease (ICD) code data | ICD data may underestimate the number of deaths related to AMR. This is because mortality statistics often allow selection of only one main underlying cause of death, which means that only the underlying condition that led to the hospital admission would be featured as cause of death, instead of other intercurrent conditions such as infections (e.g. hospital-acquired bacterial infections). |
| Mortality due to AMR infection using clinical definitions | Medical charts | Time-consuming; also, clinical definition of mortality due to AMR can be subjective. |
Estimation of excess mortality due to AMR Population attributable mortality due to AMR | A comprehensive dataset that contains individual-level clinical, microbiology and outcome data for modelling There are different study designs used to generate data, ranging from summaries of published analyses through systematic reviews to case-control studies, cohort studies and vaccine-probe trials* | Can be computationally challenging to perform analysis locally. To derive an unbiased estimate of this measurement, good quality, high-dimensional, detailed and large sample size data would often be needed; that is, representative, complete and accurate microbiology test result data, clinical data that captures pre-defined confounders, and patient treatment outcome data. |
Footnotes
*The design and analytical approach of vaccine-probe trials are beyond the scope of this course, but if you’re interested in this area you might want to refer to Feikin et al.’s review article ‘Use of vaccines as probes to define disease burden’ (2014). Examples of use of this design include for the burden of pneumococcal pneumonia disease, which remains an important bacterial infection for children in many low- and middle-income countries.Table 2 builds on the data sources and methods outlined in Table 1 and covers the corresponding analytical approaches used to interpret each measurement and their implications for estimating the health impact of AMR.
| Measurements | Analytical approaches | Interpretations |
|---|---|---|
| Crude mortality | Numerator: total counts of deaths with AMR infection in a target population over a specific time period. Denominator: total number of patients with AMR infections in the same target population over the same time period. | In some literature this is used interchangeably with the case fatality ratio. There is no causal interpretation to this measurement. Often it is challenging to compare this measurement between different settings or between different time points in the same setting; this is because difference could be due to many factors, including patient characteristics and heterogeneity in a third variable that could influence the risk of AMR infection and mortality. |
| Mortality due to AMR infection using clinical definitions | Numerator: the number of deaths that are clinically defined as directly or indirectly due to AMR infection in a target population over a specific time period. Denominator: total number of patients clinically suspected of bacterial infection and confirmed by microbiology culture that AMR pathogen(s) of interest was/were present in at least one clinical sample in the same target population over the same time period. | This is not a common measure of health impact due to AMR infections for reasons of practicality. This measurement could have an individual-level causal interpretation.* However, the clinical definition of death due to AMR infection can be subjective, which impedes the ability to compare across settings or across time. |
| Estimation of excess mortality due to AMR | Often estimated using modelling approaches | An unbiased** estimate of this measurement has a population-level causal interpretation to inform the preventable risk of mortality that would not have occurred had individuals not experienced AMR infection. This is the absolute difference between the observed risk of mortality among individuals who were factually exposed (i.e. had an AMR infection) and the probability of mortality that there would have been in the absence of AMR infection. |
| Population attributable mortality due to AMR | Often estimated using modelling approaches | An unbiased** estimate of this measurement has a population-level causal interpretation that informs the proportion of deaths that would not have occurred in the absence of AMR infection in the total population. This is often used a key parameter to estimate the number of deaths due to AMR infection. |
| Number of deaths due to AMR infection | This could be calculated by multiplying the estimated number of cases of AMR infection by excess mortality due to AMR. | An unbiased** estimate of this measurement has a population-level causal interpretation to inform the preventable number of deaths that would not have occurred had individuals not experienced AMR infection. |
Footnotes
*Please refer to the section on common terminologies for health burden of AMR in this course for detailed definition and underlying assumption of the two comparators. **It is often challenging to derive an unbiased estimate in practice and interpretation needs to carefully consider different source of biases.4.2 Interpreting estimates
